[plt-scheme] Web framework question

From: Shriram Krishnamurthi (sk at cs.brown.edu)
Date: Mon Apr 12 20:38:03 EDT 2010

Hi Johan,

I want to focus on the first issue you raise, since I think Jay has
done a good job of starting to address the rest.

I think you're both overstating and understating the continuation
issue here.  Two pertinent issues:

- Continuations do not force you into a linear program stream.  It is
quite common for continuations to represent multiple future options,
returns to previous points, and so on.  That is, they represent all
the richness of control flow that programs have.  In this sense, you
are overstating their limitations.

- We are very much concerned -- more so than just about anyone else --
with how programs behave *in the face of user interactions*,
especially those operations provided by the browser.  Indeed, we made
the point that the browser buttons and operations are essentially new
control operators in themselves, imposed from the outside on the
program by the user (and perhaps not properly imagined by the
programmer).  We provided a clean account of how one should arrange
their data relative to what kind of user interaction experience one
wanted the user to have.  In this sense, you are understating their
utility for structuring the Web computation in all its glory.

- We also realized the importance of giving meaning to these
resumption points (continuations).  Thus we have not one control
operator (SEND/SUSPEND) but four, to reflect the liveness or deadness
of the prior and future computation.

Some of this is hard to discern from our older papers, because we
thought everyone understood these issues when writing the papers and
later found that virtually nobody did.  Some things are clearer in our
journal paper (2006 or 2007), as well as in my talks, some of which
are linked from my Web page.  I will add that I don't think any other
continuation-based framework (Seaside, etc.) achieved this level of
maturity in thinking about computation and state.

A different way to put it is that you are free to not use any
continuations at all, and program in a much more traditional sense.
Sometimes, however, you DO want to arrange for a linear (or other
structured) flow of control.  For instance, maybe you need to gather
personal and billing information, you want to do it over three pages
(to keep the forms simple), and later pages should refer to earlier
pages (to customize the presented text).  This is a small
sub-computation that is perfect for structuring using SEND/SUSPEND.
The rest of your Web app does not need to be written in this style.

So, I do not see our use of continuations as a hindrance or a
straightjacket.  Rather, you are free to use the traditional style
where you want.  However, where you do need to structure a
(sub-)computation, the continuations are there to make your
computation's source code clear and hence make the structure explicit.
Combined with Jay's more recent work on the systems side of things,
this is all win, no loss.

We'd love to engage in this discussion further so we can really
understand your viewpoint and hopefully clarify some of these issues,
such that you can make the most informed choice.

Shriram


Posted on the users mailing list.